Performance of questionnaires to predict sleep‐disordered breathing in acute stroke patients

Sleep-disordered breathing is common in stroke and may negatively affect its outcome. Screening for sleep-disordered breathing in this setting is of interest but poorly studied. We aimed to evaluate the performance of eight obstructive sleep apnea screening questionnaires to predict sleep-disordered breathing in acute stroke or transient ischaemic attack patients, and to assess the impact of stroke/transient ischaemic attack-specific factors on sleep-disordered breathing prediction. We analysed acute stroke/transient ischaemic attack patients (N = 195) from a prospective cohort ("Sleep Deficiency and Stroke Outcome study"). Assessments included anthropometrics, stroke-specific parameters, sleep history, an in-hospital respiratory polygraphy within the first week after stroke, and obstructive sleep apnea screening questionnaires (Berlin Questionnaire, Epworth Sleepiness Scale, STOP-BANG, NoSAS, Sleep Apnea Clinical Score, No-Apnea, Sleep Obstructive apnea score optimized for Stroke, SLEEP-IN). In a binary classification task for respiratory event index ≥ 15 per hr, we evaluated the performance of the above-mentioned questionnaires. We used logistic regression to identify predictors for sleep-disordered breathing in this cohort. The areas under the curve for respiratory event index ≥ 15 per hr were: Berlin Questionnaire 0.60; STOP-BANG 0.72; NoSAS 0.69; No-Apnea 0.69; Sleep Apnea Clinical Score 0.75; Epworth Sleepiness Scale 0.50; Sleep Obstructive apnea score optimized for Stroke 0.58; and SLEEP-IN 0.67. The No-Apnea had the lowest false omission rate (0.13), a sensitivity of 0.97 and a specificity of 0.12. In multiple logistic regression analysis (respiratory event index ≥ 15 per hr), age, neck circumference, National Institutes of Health Stroke Scale at admission, prior stroke, cardioembolic stroke aetiology and observed apneas were associated with sleep-disordered breathing. The logistic regression model performed similar (area under the curve 0.80) to Sleep Apnea Clinical Score (p = 0.402) and STOP-BANG (p = 0.127), but outperformed the other questionnaires. Neither existing questionnaires nor our statistical model are sufficient to accurately diagnose sleep-disordered breathing after stroke, thus requiring sleep study evaluation. The No-Apnea questionnaire may help to identify patients amenable to sleep testing.